On a reasonable disaggregate criterion of population bioequivalence admitting of resampling-free testing procedures

The aggregate criterion of population bioequivalence (PBE) recommended in the new FDA guidance for in vivo bioavailability studies is criticized for using the same distributional parameter as a reference for scaling the squared distance between the averages and the difference in variability. In order to circumvent this and other difficulties inherent in the approach to PBE having been recommended thus far, we introduce a disaggregate criterion instead which requires of population bioequivalent drug formulations that both the difference in means scaled by the pooled intrasubject standard deviation be sufficiently small in absolute value, and the total variability be not substantially increased under the test as compared to the reference formulation. For testing the first of the statistical hypotheses associated with this combined criterion, an exact optimal procedure based on the ordinary two-sample t-statistic is presented. As a test for equivalence with respect to total variability, a modification of Liu and Chow's one-sided test procedure for the assessment of intrasubject variability is recommended. Both subtests are combined by means of the intersection-union principle explained here in general terms, to form an overall test for disaggregate PBE maintaining any specified significance level. The power of the combined procedure is investigated by means of Monte Carlo simulation.